Overview

TensorFlow 2.17, released on August 7, 2024, brings performance improvements for model training and inference.

Main Features

Improved performance

XLA compiler and runtime optimizations deliver significant performance gains on training and inference workloads.

python 
import tensorflow as tf

# Simple model with XLA compilation
model = tf.keras.Sequential([
    tf.keras.layers.Dense(128, activation='relu'),
    tf.keras.layers.Dense(10, activation='softmax'),
])

model.compile(
    optimizer='adam',
    loss='sparse_categorical_crossentropy',
    jit_compile=True,  # XLA compilation
)
print(f'TensorFlow {tf.__version__}')

Sources